On the genealogy of machine learning datasets: A critical history of ImageNet

By Emily L. Denton, A. Hanna, Razvan Amironesei, Andrew Smart, Hilary Nicole. 2021

220 citations recorded by the source corpus.

In response to growing concerns of bias, discrimination, and unfairness perpetuated by algorithmic systems, the datasets used to train and evaluate machine learning models have come under increased scrutiny. Many of these examinations have focused on the contents of machine learning datasets, finding glaring underrepresentation of minoritized groups. In contrast, relatively little work has been done to examine the norms, values, and assumptions embedded in these datasets. In this work, we conceptualize machine learning datasets as a type of informational infrastructure, and motivate a genealogy as method in examining the histories and modes of constitution at play in their creation. We present a critical history of ImageNet as an exemplar, utilizing critical discourse analysis of major texts around ImageNet’s creation and impact. We find that assumptions around ImageNet and other large computer vision datasets more generally rely on three themes: the aggregation and accumulation of more data, the computational construction of meaning, and making certain types of data labor invisible. By tracing the discourses that surround this influential benchmark, we contribute to the ongoing development of the standards and norms around data development in machine learning and artificial intelligence research.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Safety Research, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Safety Research

In Evaluation & Benchmarks

Citations in the corpus

AI and the Everything in the Whole Wide World Benchmark (2021, background): “All datasets come with an embedded perspective — there is no neutral or universal dataset [Haraway, 1988, Stitzlein, 2004, Gebru, 2020, Denton et al., 2021, Scheuerman et al., 2021].”
Vector-quantized Image Modeling with Improved VQGAN (2021, background): “Nevertheless, it is important to recognize that ImageNet itself was created with biases in terms of image selection and label annotation as a result of its process of creation (Denton et al., 2021).”
Scaling Autoregressive Models for Content-Rich Text-to-Image Generation (2022, background): “Since large text-to-image models are foundation models [83] — enabling both a range of system applications as well as finetuning for specific image generation tasks — they act as a form of infrastructure which shapes our conceptions of what is both possible and desirable [84, 85].”
Multimodal datasets: misogyny, pornography, and malignant stereotypes (2021, background): “Various works [6, 7, 8, 9] have highlighted gender, racial, and geographical biases surrounding the sourcing of image datasets as well as the opacity of such endeavors [10].”
Data Governance in the Age of Large-Scale Data-Driven Language Technology (2022, background): “Very recent research — most published within the last year — has begun to analyze dataset values [18, 40], question assumptions around dataset use [100], unpack what is represented in ML datasets [43, 88, 124] and establish basics of how an organized dataset lifecycle might proceed [74].”
Power to the People? Opportunities and Challenges for Participatory AI (2022, background): “One significant adoption of non-expert participation was in the construction of large scale benchmark datasets such as ImageNet [22], where the research team utilized over 49,000 workers from Amazon’s mechanical turk (Mturk) platform across 167 countries [23] to perform image recognition tasks to fi”
A Framework for Deprecating Datasets: Standardizing Documentation, Identification, and Communication (2021, methodology): “the data creation stage via checklists [58] and data sheets [34] to the study of dataset usage and genealogy [32, 50], licensing and auditing [11, 18, 78, 88], as well as data stewardship and maintenance [72, 74, 102].”
Data Augmentation in Classification and Segmentation: A Survey and New Strategies (2023, background): “Note that some of these datasets, especially ImageNet, are considered “big data” [36] and may not require data augmentation techniques to further increase their size.”
Participation Is not a Design Fix for Machine Learning (2020, background): “labels [5], that they were collected without adequately informed consent [17], and that they are imbalanced across demographic groups in ways that lead to bias in ML systems trained on such datasets [69].”
Can Machines Help Us Answering Question 16 in Datasheets, and In Turn Reflecting on Inappropriate Content? (2022, background): “…for language models, for instance, models producing stereotypical and derogatory content [Bender et al. 2021], and for vision model respectively CV datasets highlighting, e.g., gender and racial biases [Denton et al. 2021; Larrazabal et al. 2020; Steed and Caliskan 2021; Wang et al. 2020].”
Cultural Incongruencies in Artificial Intelligence (2022, background): “…play a critical role in the development of AI systems—e.g., through “labels” on data, and how “knowl-edge”, “objectivity” [5], “reality/truth” [24], and “system objectives” are constructed—the cultural norms of the AI developers and researchers also pervasively infuse the AI systems [10, 6, 23].”